2. and nongeneric statements (Cimpian & Cadena, 2010; Cimpian &
Markman, 2009, 2011). Although whether a statement refers to a kind
is not overtly relevant to the causal source of the facts described in it,
children seem to use this feature of a statementās meaning to make
inferences about underlying causes. These causal inferences or attri-
butions, in turn, tend to change how children understand the facts they
learn.
Since children acquire much of their knowledge through language
(e.g., Gelman, 2009; Harris, 2002), the impact of the generic/
nongeneric distinction on childrenās causal understanding of the
world may be extensive. The fact that generic statements are com-
monplace in child-directed speech (Gelman, Chesnick, & Waxman,
2005; Gelman, Coley, Rosengren, Hartman, & Pappas, 1998;
Gelman, Goetz, Sarnecka, & Flukes, 2008; Gelman & Tardif, 1998;
Gelman, Taylor, & Nguyen, 2004), along with childrenās early de-
veloping ability to comprehend them (Cimpian & Markman, 2008;
Cimpian, Meltzer, & Markman, in press; Gelman & Raman, 2003),
further underscores the potential importance of the relationship be-
tween this linguistic distinction and childrenās causal attributions.
We begin with a brief review of the existing evidence suggest-
ing that the generic/nongeneric distinction shapes childrenās attri-
butions. We then describe a potential mechanism by which generic
and nongeneric statements affect childrenās causal attributions and
test two key predictions of this account.
Generics and Causal Attributions: The Evidence
When learning about novel features of living things, 4- and 5-year-
old children are more likely to attribute functional, life-sustaining
powers to these features if they are introduced via generic than via
nongeneric statements (Cimpian & Markman, 2009). For example,
when children were told that snakes have holes in their teeth, they
explained the presence of the holes by invoking the functions they
could fulfill: allowing snakes to āswallow things,ā or to āchew better,ā
or to ādrink the blood out of predators.ā In contrast, framing this
feature as characteristic of a particular snake led to a different pattern
of causal attributions, dominated by appeals to prior, often accidental,
causes. For example, one child thought the holes were caused by bug
bites (āA bug came in its room, and it bited his teethā), while another
thought they were caused by food (āMaybe itās because it eats so
much little stuffā).
Cimpian and Cadena (2010) extended this conclusion to the do-
main of artifacts. When features of unfamiliar artifacts were intro-
duced via generic statements, 5-year-olds often conceptualized them
as functional aspects of the artifactsā intended design. For example,
when they were told that dunkels are sticky, children often believed
that dunkels were actually made that way (sticky), and they attributed
the stickiness to a function dunkels were meant to perform (e.g., they
are sticky āso that drinks donāt fall and stay onā). When the same
properties were framed nongenerically (e.g., āThis dunkel is stickyā),
their meaning changed. For example, children understood the sticki-
ness not in terms of functions but in terms of prior mechanistic causes
other than the creatorās intent (e.g., āācause the dog chewed on itā).
Novel information about other people is also understood differently
depending on the generic/nongeneric format in which it is introduced
(Cimpian & Markman, 2011). For example, when an ability is said to
characterize an entire social kind (e.g., āBoys are good at a game
called āgorpāā), preschoolers think of it as stemming from deep,
essential traits of the category (e.g., āācause boys grow up fast,ā
āmaybe ācause they are stronger than girlsā). This ability is thus
understood in quasibiological terms, as part of boysā inherent endow-
ment. In contrast, when the same ability is introduced via a nongeneric
statement (e.g., āThereās a boy who is good at a game called āgorpāā),
preschoolers attribute it to effort and practice much more frequently
than in the generic case (e.g., ābecause he practiced a lot of times,ā
āmaybe he learned at schoolā). By promoting effort attributions, the
nongeneric context leads to an understanding of this novel ability as
the malleable outcome of a process that is under childrenās control
rather than as a fixed biological reality.
The Present Research
In what follows, we propose a partial account of the process by
which the generic/nongeneric distinction shapes childrenās attribu-
tions. The central claim is that childrenās causal attributions are
determined not by whether the statements being explained are
generic or nongeneric but instead by whether the beliefs relevant to
these statements are generic or nongeneric. These beliefs may
consist of information previously stored in semantic memory and
activated by the current statements, or they may be constructed in
the moment on the basis of the statementsā meaning, the context,
childrenās naive theories, and so on. Either way, we hypothesize
that it is the generic/nongeneric format of these beliefs, and not the
generic/nongeneric format of the language per se, that is the
proximal cause of childrenās attributions.
Of course, the genericity of a statement and the genericity of the
corresponding beliefs often coincide. For example, a child who
was exposed to a novel generic statement such as āGirls are good
at a game called ātookiāā would probably also form a generic belief
that girls, as a kind, are good at tooki. Thus, any explanations that
the child generated for girlsā tooki ability in this context would not
allow researchers to distinguish between the belief-proximal ac-
count we proposed and the alternative, language-proximal account
(according to which the generic/nongeneric format of the state-
ments being explained is the proximal cause of childrenās attribu-
tions): Because the language and the resulting belief are both
generic in this context, both accounts predict that children would
attribute girlsā tooki ability to their inherent talents (Cimpian &
Markman, 2011). Thus, in order to adjudicate between these two
accounts, it was necessary to create a task context in which the
language and the corresponding beliefs differed in genericity. To
illustrate, imagine that the child who was told that girls are good
at tooki (and formed the corresponding generic belief) later heard
a statement about a particular child, such as āJane is good at
tooki.ā Although this statement is nongeneric, it is consistent with
the prior generic belief that girls are good at tooki and is thus most
likely assimilated under its scope.1
Thus, on the belief-proximal
account, Janeās ability should be attributed to an inherent talent
because the causal attribution process operates on the resulting
belief, which in this case is generic. In contrast, on the language-
proximal account, the causal attribution process operates locally,
1
In contrast, a generic statement (e.g., āGirls are good at tookiā) would
most likely cause a prior nongeneric belief (e.g., that Jane and Katie are
good at tooki) to be updated to generic format. However, because both the
statement under consideration (āGirls are good at tookiā) and the resulting
belief are generic here, this context cannot differentiate between the
language-proximal and belief-proximal accounts.
160 CIMPIAN AND ERICKSON
3. on the meaning of the nongeneric statement about Janeās tooki
ability, and thus the more probable attribution would be to effort
and practice. We elaborate this pair of contrasting predictions in
the next section. We then go on to derive a second set of predic-
tions that distinguish between the belief-proximal and the
language-proximal accounts.
Predictions
The first prediction of the belief-proximal account: Generic
statements affect attributions even in nongeneric contexts.
According to the belief-proximal hypothesis, the factor that deter-
mines childrenās causal attributions for a property is the genericity
of childrenās beliefs on the topic (e.g., whether they believe that
math ability is a feature of boys in general or a feature of particular
boys) and not the genericity of the statement in which this property
is heard at the particular point in time when the attribution is
generated (e.g., whether children hear āBoys are good at mathā or
āJohnny is good at mathā). If this hypothesis is correct, then one
would expect a novel generic statement such as āGirls are good at
a game called ātookiāā to leave its mark, via the generic belief it
induces, on childrenās causal attributions in a broad range of
subsequent contextsāmore precisely, in any contexts that fall
under the scope of the represented generic belief. In other words,
features learned from generic statements such as āGirls are good at
tookiā should be attributed to essential factors not just in contexts
where children are reasoning about these statements per se or
about the kind generalizations to which they refer. According to
our belief-proximal hypothesis, even circumstances that involve
specific individuals displaying these features (e.g., seeing a par-
ticular girl doing well at tooki, or hearing āJane is good at tookiā)
may lead to such essentialized attributions because these circum-
stances are likely to activate the relevant generic beliefs from
semantic memory. This is, in fact, our first prediction: that the
generic beliefs induced by generic statements will affect childrenās
attributions when they later have to reason about particular cases
that match the content of these beliefs.
To elaborate, we hypothesized that when children encounter a
situation or event that can be subsumed under an existing generic
belief, they will understand what has occurred through the lens of
this belief. That is, they will conceptualize the situation as a
particular instance of the relevant generic pattern and thus as the
product of the same (essential, inherent) causes that are responsible
for this pattern. For example, the generic belief that girls are good
at tooki will color childrenās understanding of any particular girlās
success and will lead them to attribute it to her natural abilities
rather than to hard work more often than they would have in the
absence of a generic belief about girlsā tooki ability.
Prior evidence suggesting that generic beliefs are robust is
relevant to this prediction. Most notably, the genericity of a fact
seems to be an integral aspect of its representation in long-term
semantic memory, even for 3-year-olds (Gelman & Raman, 2007;
see also Cimpian et al., in press). Generic beliefs are also partic-
ularly resistant to counterevidence (e.g., Abelson & Kanouse,
1966; Chambers, Graham, & Turner, 2008; Cimpian, Brandone, &
Gelman, 2010), which adds to their staying power and thus to their
potential ability to influence childrenās causal inferences down the
line.
To investigate this first prediction, we devised a two-step task.
In Step 1, we introduced a novel piece of information in either
generic (e.g., āBoys/Girls are really good at a game called ātookiāā)
or nongeneric (e.g., āThereās a boy/girl who is really good at a
game called ātookiāā) format. In Step 2, which was identical for all
children, the experimenter showed children a picture of a single
child and said, āThis boy/girl is also really good at tooki.ā Children
were then asked to explain this nongeneric fact, and their expla-
nations were used to determine what causal attributions they made
for it. This second step was designed to unconfound the genericity
of the beliefs induced in Step 1 from the genericity of the state-
ments that children were asked to explain. If generic statements
license generic beliefs that in turn affect the causal attributions
generated in subsequent nongeneric contexts that fall under the
scope of these beliefs, then children who heard generics in Step 1
should be more likely to attribute the ability of the individual in
Step 2 to deep, essential traits (e.g., he/she is good because he/she
is talented) than children who heard nongenerics in Step 1. On the
language-proximal account, however, childrenās attributions
should be identical across the generic and nongeneric conditions:
On this account, the factor that determines childrenās attributions
is whether the statement currently under consideration is generic or
nongeneric, and in Step 2 all children were asked to explain the
same nongeneric statement.
To clarify some terminology, our use of the terms deep and
essential is rooted in the theory of psychological essentialism (e.g.,
Bloom, 2004; Gelman, 2003; Medin & Ortony, 1989). An impor-
tant claim of this theory is that the features of a concept differ in
terms of their depth or centrality. Although there is no universally
accepted account of what makes one feature more central than
another, a plausible hypothesis is that a featureās depth/centrality is
a function of the number of other features it causes or otherwise
constrains (e.g., having a Y chromosome is a deeper feature than
having a prominent Adamās apple; Ahn, Kim, Lassaline, & Den-
nis, 2000; Medin & Ortony, 1989; but see Rehder & Hastie, 2001).
The most central feature is the category essence, the ultimate
causal source of all typical features of a category. Although there
is a single essence, features on the deep/central side of the con-
tinuum are often known as essential (e.g., Gelman, 2003; Gelman
& Wellman, 1991). These essential features are not the essence per
se, but they are essence-like in that they are judged to be important
for category membership (e.g., Gelman & Wellman, 1991; Keil,
1989) and to be causally responsible for many peripheral proper-
ties (Ahn et al., 2000, 2001; Gelman & Wellman, 1991).
Thus, our first prediction is that childrenās explanations for
nongeneric facts that fall under the scope of prior generic beliefs
will more often invoke features that are deep and essentialāin the
previously described senseāor will otherwise reveal that children
understood these nongeneric facts as being deep and essential.
The second prediction of the belief-proximal account: Ge-
neric statements are not unique in their effect on causal attri-
butions. Our belief-proximal account also motivates the follow-
ing prediction: If it is the generic or nongeneric format of the
beliefs that drives causal attributions, then any statement that
induces generic beliefs should lead to essentialized attributions,
regardless of whether the statement itself is generic or nongeneric.
Although generic statements are certainly able to induce generic
beliefs, they may not be unique in this respect. Specifically, some
nongeneric statements (e.g., āMost girls are good at tookiā) may
161GENERICS AND CAUSAL ATTRIBUTIONS
4. also license generic beliefs, which should in turn give rise to
essentialized attributions.
This prediction is supported by recent evidence that nongeneric
statements quantified with āmostā and āallā are often (mis)inter-
preted as expressing generalizations about kinds (Hollander,
Gelman, & Star, 2002; Khemlani, Leslie, Glucksberg, & Rubio-
Fernandez, 2007; Leslie & Gelman, 2011). For example, Leslie
and Gelmanās recent data suggest that both preschool-age children
and adults consistently misremember āmostā statements (e.g.,
āMost cats sweat through their pawsā) as generic (e.g., āCats sweat
through their pawsā), but they very seldom make the opposite
mistake, namely, misremembering generics as āmostā statements.2
There are at least two possible reasons for such a result. First,
unrestricted generalizations to kinds may be a default, or ācogni-
tively primitive,ā mode of generalization (Leslie, 2007, 2008).
Therefore, people may often fall back on this default generic mode
(and thus arrive at generic beliefs) when processing the more
ācognitively sophisticatedā generalizations conveyed by nonge-
neric statements quantified with āmostā and āall.ā Second, the fact
that many members of a kind display a certain property, as implied
by a sentence such as āMost girls are good at tooki,ā may license
an inductive inference translating this quantificational fact into a
fact about the relevant concept as a whole (Cimpian, Brandone, &
Gelman, 2010; Jƶnsson & Hampton, 2006). That is, people may
use the high prevalence of a feature among the entities that fall
under a concept (e.g., the high prevalence of tooki ability among
girls) to infer that this feature may be legitimately considered part
of this conceptās meaning (e.g., part of what it means to be
female). For example, adults often consider the fact that a feature
is relatively common within a kind (e.g., 50% of lorches have
purple feathers) as sufficient grounds for adopting the correspond-
ing generic belief (e.g., that lorches, as a kind, have purple feath-
ers; Cimpian, Brandone, & Gelman, 2010). On this second ac-
count, peopleās adoption of a generic belief is not a mistake but
rather a legitimate outcome of the interpretive process.
The claim here is not that generic and wide-scope quantified
statements are synonymous. There is no doubt that they have
distinct linguistic meanings (e.g., Carlson, 1977; Cimpian, Bran-
done, & Gelman, 2010; Cimpian, Gelman, & Brandone, 2010;
Leslie, 2007, 2008). Rather, what we are claiming is that this
distinction, which is very clear-cut at the level of the statements
themselves, becomes less straightforward at the level of the beliefs
that these statements ultimately lead people to adopt. In particular,
wide-scope quantified statements may often end up giving rise to
generic beliefs, either because their interpretation is biased by a
default kind-based mode of generalization (Leslie, 2008) or be-
cause they license additional inductive inferences about the con-
ceptās meaning. Either way, once a generic belief is formed, the
essentialist causal attributions should followāthis is what the
belief-proximal account would predict.
We tested this prediction in the context of the two-step task
described earlier by comparing the generic and nongeneric condi-
tions to a third condition in which the novel properties were
introduced via statements quantified with āmostā3
(e.g., āMost
boys/girls are really good at a game called ātookiāā). If the āmostā
statements in Step 1 lead people to form generic beliefs, then the
nongeneric facts in Step 2 (e.g., āThis boy/girl is also really good
at tookiā) should be attributed to deep, essential causes. In other
words, we predicted that the causal attributions of participants in
the āmostā condition would be similar to those of participants in
the generic condition and different from those of participants in the
nongeneric condition. Again, it is important to point out that this
prediction is contrary to the language-proximal account: Because
in Step 2 all participants were asked to explain a fact conveyed in
a nongeneric statement and because on the language-proximal
account the causal attribution process operates at the level of the
statements that are being explained, this account would predict no
differences in causal attributions across the three conditions.
Summary
We derived two sets of predictions that distinguish the belief-
proximal and the language-proximal accounts. First, the belief-
proximal, but not the language-proximal, account predicts that the
generic beliefs formed on the basis of generic statements will
influence childrenās causal attributions in subsequent nongeneric
contexts that fall under their scope. Specifically, children should
attribute the nongeneric facts in Step 2 to essential causes more
often when these facts are linked to existing generic beliefs (ge-
neric condition) than when they are not (nongeneric condition).
Second, the belief-proximal, but not the language-proximal, ac-
count predicts that any statements that lead to the formation of
generic beliefs, regardless of whether these statements are them-
selves generic or not, should similarly bias childrenās attributions
towards essential causes. In light of the evidence that āmostā-
quantified statements may often license generic beliefs (e.g., Leslie
& Gelman, 2011), the belief-proximal account predicts that such
statements should be similar to generic statements in their likeli-
hood of inducing an essentialized causal perspective on the non-
generic facts in Step 2. To the extent that these two predictions are
confirmed, they would support the belief-proximal account and
undermine the language-proximal account.
Method
Participants
The participants belonged to three age groups: 4- and 5-year-
olds (n Ļ 73; 37 girls and 36 boys; mean age Ļ 5 years; range Ļ
from 4 years to 5 years 11 months), 6- and 7-year-olds (n Ļ 75; 38
girls and 37 boys; mean age Ļ 7 years; range Ļ from 6 years to
7 years 11 months), and undergraduates (n Ļ 72; 37 females and
35 males). An additional 17 children were tested but excluded
from the sample. We used two criteria for determining whether
2
It is unlikely that this result was simply due to poor memory for the
first word of the quantified statements: Participants seldom misrecalled
āsomeā-quantified statements (e.g., āSome cats . . . ā) as genericācontrary
to what would be expected if they were simply dropping the first word of
the original sentences. Similarly, a separate control study found that
children were unlikely to forget the first word of quantified statements such
as āNo bees have six eyesā (i.e., they seldom recalled these statements as,
e.g., āBees have six eyesā). In sum, the tendency to misremember āmostā
statements as generic was in all likelihood driven by the beliefs that these
statements ultimately led people to adopt.
3
Although āmostā-quantified statements are also nongeneric, for sim-
plicity we will refer to only the nongeneric statements about a single
individual (e.g., āThereās a boy/girl . . . ā) as ānongeneric.ā
162 CIMPIAN AND ERICKSON
5. a child should be excluded: (a) the child made no response, said
āI donāt know,ā or gave unintelligible responses for the first
three trials (testing was stopped at this point); and (b) the child
made no response, said āI donāt know,ā or gave unintelligible
responses for half or more of the trials over the course of the
entire session. The present exclusion rate (10.3%) is well within
the range observed in other studies that required young children
to generate open-ended explanations (Cimpian & Cadena, 2010;
Cimpian & Markman, 2009, 2011).
The children were recruited in a small city in the midwestern
United States, either from local preschools and elementary schools
or from a database of families interested in participating in devel-
opmental studies. The undergraduates, who were included to in-
vestigate potential developmental differences in the influence of
generics, were recruited from the subject pool at a large public
university and received course credit for participation. Although
demographic information was not collected formally, the partici-
pants were mostly European American and represented a range of
socioeconomic backgrounds.
Materials and Design
The information presented to participants consisted of eight
novel properties. Four of these properties were about abilities (e.g.,
being good at a game called ātookiā), and the other four were about
biological features (e.g., having something called āthromboxaneā
in oneās brain; see Table 1 for the complete set of properties). The
ability and biological feature trials were presented in two separate
four-trial blocks, and the order of these blocks was counterbal-
anced across children. Participants were randomly assigned to one
of three wording conditions, which differed only in the wording
used to introduce the eight properties: nongeneric, generic, or
āmostā (see Table 1).
The design can be summarized as follows: 3 (age group: 4- and
5-year-olds vs. 6- and 7-year-olds vs. adults; between subjects) Ļ«
2 (property type: abilities vs. biological features; within subject) Ļ«
3 (wording: nongeneric vs. generic vs. āmostā; between subjects).
A number of other variables were counterbalanced across par-
ticipants: the order of the eight properties, the gender with which
each property was paired, and the gender of the first trial. The
gender of the eight trials alternated, such that half of the trials
within each block were about girls and half were about boys.
Procedure
Children were tested individually in a quiet room in their school
or in the lab. Undergraduate participants were tested in groups on
a paper-and-pencil version of this task. For children, the experi-
menter wrote down childrenās responses during testing, but the
sessions were also videotaped for later transcription. To motivate
children, the experimenter introduced them to a stuffed animal that
was ātrying to figure some things outā and asked them to help it.
Each of the eight trials proceeded in two steps, as detailed next.
Step 1: Introducing the Information. In this step, children
heard the experimenter introduce a novel piece of information in
one of three formats, depending on the wording condition to which
they had been assigned: (a) nongeneric format (e.g., āI wanna tell
you something interesting about a boy/girl. Thereās a boy/girl who
is really good at a game called ātookiāā), (b) generic format (e.g.,
āI wanna tell you something interesting about boys/girls. Boys/
Table 1
The Items Used
Abilities Biological features
Game Brain
NG: Thereās a boy/girl who is really good at a kind of game called
ātooki.ā
NG: Thereās a boy/girl who has something called āthromboxaneā
in his/her brain.
G: Boys/Girls are really good at a kind of game called ātooki.ā G: Boys/Girls have something called āthromboxaneā in their
brains.
Most: Most boys/girls are really good at a kind of game called
ātooki.ā
Most: Most boys/girls have something called āthromboxaneā in
their brains.
Sport Bones
NG: Thereās a boy/girl who is very good at a kind of sport called
āleeming.ā
NG: Thereās a boy/girl who has these things called āosteoclastsā
in his/her bones.
G: Boys/Girls are very good at a kind of sport called āleeming.ā G: Boys/Girls have these things called āosteoclastsā in their bones.
Most: Most boys/girls are very good at a kind of sport called
āleeming.ā
Most: Most boys/girls have these things called āosteoclastsā in
their bones.
Dance Blood
NG: Thereās a boy/girl who is really good at a kind of dance called
āludino.ā
NG: Thereās a boy/girl who has something called āfibrinogenā in
his/her blood.
G: Boys/Girls are really good at a kind of dance called āludino.ā G: Boys/Girls have something called āfibrinogenā in their blood.
Most: Most boys/girls are really good at a kind of dance called
āludino.ā
Most: Most boys/girls have something called āfibrinogenā in their
blood.
Puzzle Muscles
NG: Thereās a boy/girl who is very good at a kind of puzzle called
āzool.ā
NG: Thereās a boy/girl who has these things called āsarcomeresā
in his/her muscles.
G: Boys/Girls are very good at a kind of puzzle called āzool.ā G: Boys/Girls have these things called āsarcomeresā in their
muscles.
Most: Most boys/girls are very good at a kind of puzzle called
āzool.ā
Most: Most boys/girls have these things called āsarcomeresā in
their muscles.
Note. āGā stands for āgeneric,ā and āNGā stands for ānongeneric.ā
163GENERICS AND CAUSAL ATTRIBUTIONS
6. Girls are really good at a game called ātookiāā), or (c) āmostā
format (e.g., āI wanna tell you something interesting about most
boys/girls. Most boys/girls are really good at a game called
ātookiāā). The key property was repeated once in the appropriate
format, and then the experimenter proceeded to Step 2 for that
property.
Step 2: Assessing Childrenās Attributions for a Subsequent
Nongeneric Fact. This step was identical for all children, re-
gardless of wording condition. Children were shown a picture of a
single child (whose gender matched that used in Step 1) and were
then told that this child possesses the property introduced in Step
1. For example, on one trial the experimenter said, āAnd hereās
another boy/girl. And you know what? This boy/girl is also really
good at tooki. He/She is also really good at tooki.ā It was partic-
ularly important to refer to the child in the picture as āanother
boy/girlā in the nongeneric condition, so as to differentiate this
child from the child talked about in Step 1. For consistency, we
used this phrase in the generic and āmostā conditions as well.
Next, children were asked to explain this nongeneric fact: for
example, āWhy do you think that is? Why is this boy/girl really
good at this game called ātookiā?ā After writing down childrenās
answer, the experimenter proceeded to introduce the next property
(Step 1). If children said they did not know, they were reassured
that there were no right or wrong answers and asked to make a
guess. If they still said they did not know, the experimenter went
on to the next trial but returned to the unanswered item at the end
of the session for a final try. (Only 5.6% of trials required such a
return.) Eight different pictures of boys and girls were used during
Step 2, one for each property.
At the end of the session, children in the generic and āmostā
conditions received a short debriefing, in which it was explained to
them that āin real lifeā boys and girls are good at the same things
and have essentially the same things inside their bodies.
Coding
We coded participantsā explanations into two main categories
(for a similar coding scheme, see Cimpian & Markman, 2011).
First, responses revealing that the to-be-explained property was
understood as a natural, deep, inherent fact about the child in the
picture were coded as essentialized. For example, one child ex-
plained the presence of fibrinogen in the blood of the child in the
picture by saying that āshe was made like that from God.ā Second,
responses revealing that the to-be-explained property was under-
stood as the result of a mechanistic, often externally driven causal
process were coded as nonessentialized. For example, one child
thought that the child in the picture was good at leeming ābecause
his dad teached him.ā Additional examples of essentialized and
nonessentialized explanations are provided in Table 2.
Aside from the essentialized and nonessentialized categories,
the coding scheme also included a syllogistic reasoning category.
For example, if a participant was told that girls are good at tooki
(Step 1) and was then asked why a particular girl is good at tooki
(Step 2), a syllogistic explanation would be that she is good
because girls are good, and she is a girl. In principle, such expla-
nations could reflect an essentialized understanding of the infor-
mation. For instance, explaining why a girl is good at tooki by
saying ābecause she is a girlā seems to imply that this property is
a natural, inherent aspect of who she is. In the context of our task,
however, a participant might also arrive at this explanation by
simply integrating all of the information provided, without any
further commitments as to whether this property is essential. This
interpretive ambiguity led us to decide against counting syllogistic
explanations as essentialized. This was a conservative decision, as
syllogistic explanations occurred almost exclusively in the generic
and āmostā conditions (see Table 3), where we predicted essen-
tialized explanations would be most frequent. Also note that,
Table 2
Examples of Essentialized and Nonessentialized Explanations
Essentialized explanations Nonessentialized explanations
Inherent explanations Problem explanations
Description: These explanations imply that the property is a natural or
normal aspect of development.
Description: These explanations imply that the property is the result
of a problem, accident, injury, or disease.
Examples
āShe was made like that from Godā [blood]
āThatās the way bones areā [bones]
āBecause thatās the way of the worldā [sport]
Trait explanations
Description: These explanations imply that the property is the result
of some intrinsic trait or preference of the individual.
Examples
āHeās strongā [game]
āBecause she has good legsā [dance]
āBecause heās really smartā [puzzle]
Functional explanations
Description: These explanations imply that the property serves
some (usually life-sustaining) function.
Examples
āBecause it helps your bodyā [blood]
āSo her muscles can get strongā [muscles]
āSo she can walk or somethingā [bones]
Examples
āBecause something got on him, maybe a bug, that might have
sucked his bloodā [blood]
āSheās sickā [muscles]
āHer bones arenāt tight enoughā [bones]
Practice explanations
Description: These explanations imply that the property developed
through practice, learning, instruction, or exercising.
Examples
āShe practiced every dayā [puzzle]
āBecause his dad teached himā [sport]
āShe took lessonsā [dance]
External explanations
Description: These explanations imply that the property is the result
of some situational or external cause (other than contagion/
infection or instruction).
Examples
āBecause he lives at Kentuckyā [game]
āSheās been eating vegetables a lotā [muscles]
āHe has the jacket for itā [dance]
Note. The word in square brackets indicates the item for which the explanation was provided.
164 CIMPIAN AND ERICKSON
7. within the generic and āmostā conditions, adults were more likely
to produce syllogistic explanations than children were. This age
difference may be due to adultsā superior formal reasoning skills
(e.g., Galotti, 1989) or to the fact that adultsā booklets provided
simultaneous access to the information in Steps 1 and 2, which
may have helped them detect the syllogistic structure of this
information.
Any responses that did not fit into the three categories were
coded as other. For example, one child explained why a girl is
good at the ludino dance by saying, āBecause she is my best friend,
and I love her,ā and another child explained why a boy had
sarcomeres in his muscles by saying, āBecause he knew he had it.ā
These explanations decreased in frequency with age (M4 ā5 Ļ 2.33
vs. M6 ā7 Ļ 1.21 vs. Madult Ļ 0.94 on eight trials) but did not show
any condition differences (Mnongeneric Ļ 1.47 vs. Mgeneric Ļ 1.49
vs. Māmostā Ļ 1.53), so we will not discuss them further.
Coding was blind to the wording of the property in Step 1 and
followed a present/absent rule: Any explanation categories that
were present on a particular trial were assigned a 1; all absent
categories were assigned a 0. To assess reliability, a second re-
searcher coded 180 of the 220 transcripts. Intercoder agreement
was 91.9% (ā¬ Ļ .82) for the essentialized explanations, 93.8%
(ā¬ Ļ .86) for the nonessentialized explanations, and 98.8% (ā¬ Ļ
.96) for the syllogistic explanations. Disagreements were resolved
by discussion.
Data Analysis
Our main analyses compared the distribution of essentialized
and nonessentialized explanations across wording conditions.
However, we did not compare the raw numbers of these explana-
tions. This decision was motivated by the condition differences in
the number of syllogistic explanations: These explanations clus-
tered in the generic and āmostā conditions and were rare in the
nongeneric condition (see Table 3). The substantial presence of
syllogistic explanations in the generic and āmostā conditionsāin
combination with subjectsā tendency to generate a single explana-
tion per trial (they did so on over 93% of trials)ādrove down the
raw frequency of all other explanation types in these two condi-
tions. This pattern was most pronounced in the adultsā data, where
syllogistic explanations accounted for roughly half of all explana-
tions in the generic and āmostā conditions, leaving less room for
other responses.
As a consequence, we used as our dependent measures the
proportions of nonsyllogistic explanations that were essentialized
and nonessentialized. This measure corrected for the imbalance
due to the syllogistic explanations, and thus reduced the likelihood
that comparisons across wording conditions and across age groups
would be biased by this imbalance. Participants who produced
only syllogistic explanations on all eight trials were dropped from
the analysis (three 4- to 5-year-olds, two 6- to 7-year-olds, and 21
undergraduates; all had been in the generic and āmostā conditions).
The distribution of these dependent measures departed from
normality (ShapiroāWilk test ps Ļ½ .001), violating one of the main
assumptions underlying parametric statistics. Therefore, we ana-
lyzed participantsā responses with nonparametric repeated-
measures ordinal logistic regressions (RM-OLRs) computed using
the Generalized Estimating Equations command in SPSS 17 (for
analogous analyses, see Cimpian & Cadena, 2010; Cimpian &
Markman, 2009, 2011).4
Finally, note that the essentialized and the nonessentialized
proportion measures are not redundant. Statistically, the two are
not redundant because they do not make up the totality of the
nonsyllogistic explanations (recall that the explanations coded as
other were also nonsyllogistic); therefore, these measures are not
perfectly negatively correlated. In fact, the negative relationship
between the essentialized and nonessentialized measures was only
moderate in strength, r Ļ ĻŖ.62. Along the same lines, participants
were free to produce neither or both of these explanations on a
single trial; these outcomes occurred on 19.9% of the trials in-
cluded in our analyses. There are also some theoretical reasons to
suspect that these measures may not always be mirror images of
each other. For example, preschoolers are typically less likely than
older children and adults to interpret other individualsā behaviors
in terms of stable traits (e.g., Rholes & Ruble, 1984), which may
lower the overall number of essentialized explanations they pro-
duce for the features of the individual children in Step 2. In turn,
this might make it more difficult to detect the predicted differences
between wording conditions on the essentialized measure than on
the nonessentialized measure, especially in young childrenās re-
sponses. (To preview, our data confirmed this suspicion.) For the
reasons just laid out, we report the results for both of these
explanation measures.
Results
Our belief-proximal proposalāthat causal attributions are
driven not by the generic/nongeneric format of the statements
being explained but rather by the generic/nongeneric format of the
relevant beliefsāled to two specific predictions. The first predic-
tion was that, in the generic condition, participantsā causal under-
standing of the nongeneric facts in Step 2 should be filtered
through, or skewed by, the generic beliefs established in Step 1. As
a result, these participants should produce more essentialized ex-
planations and fewer nonessentialized explanations in Step 2 than
participants in the nongeneric condition. The second prediction
was that, in the āmostā condition, participantsā attributions for the
nongeneric facts in Step 2 would also be skewed towards essential
causes, arguably because āmostā-quantified statements such as the
ones in Step 1 often give rise to generic beliefs (e.g., Leslie &
4
Although not reported here, a parallel set of analyses of variance
(ANOVAs) revealed the same main findings.
Table 3
Mean Number of Syllogistic Explanations (and SDs) by Wording
Condition and Age Group
Age group
Nongeneric
condition
Generic
condition
āMostā
condition
M SD M SD M SD
4- to 5-year-olds 0.00 0.00 1.00 2.50 0.67 1.81
6- to 7-year-olds 0.12 0.43 1.08 2.58 0.58 1.35
Adults 0.42 0.88 5.42 3.34 4.54 3.59
Note. All means are out of a possible 8.
165GENERICS AND CAUSAL ATTRIBUTIONS
8. Gelman, 2011). As a result, the responses of participants in the
āmostā condition should resemble those of participants in the
generic condition and should differ from those of participants in
the nongeneric condition.
Essentialized Explanations
An RM-OLR was performed on the essentialized explanation
data, with wording condition, age group, and property type as
independent variables. Crucially, this analysis revealed a signifi-
cant main effect of wording condition, Wald ā¹2
Ļ 33.27, df Ļ 2,
p Ļ½ .001.
To test our first prediction, we performed a planned follow-up
RM-OLR comparing the generic and nongeneric conditions. As
predicted by the belief-proximal account, the participants who
heard generic statements in Step 1, and presumably formed the
corresponding generic beliefs, produced more essentialized expla-
nations (e.g., she is good at tooki because she is smart) for the facts
in Step 2 than did the participants who heard nongeneric state-
ments in Step 1, Mgeneric Ļ .55 vs. Mnongeneric Ļ .32, p Ļ½ .001.
To test our second prediction, we performed planned follow-up
RM-OLRs comparing the āmostā condition with the other two.
These analyses revealed the predicted pattern: The proportion of
essentialized explanations in the āmostā condition (M Ļ .48) was
not significantly different from that in the generic condition (M Ļ
.55), p Ļ .253, but was significantly higher than that in the
nongeneric condition (M Ļ .32), p Ļ½ .001. For example, partici-
pants who were told in Step 1 that most boys or girls are good at
a certain activity were likely to attribute the ability of the individ-
ual boys or girls in Step 2 to their natural talent or other inherent
traits, much as participants who heard generics in Step 1 did. To
reiterate, these condition differences are consistent with the belief-
proximal account but not with the language-proximal account,
since the wording of the statements children were asked to explain
(in Step 2) was identical, and nongeneric, across all three condi-
tions.
The main RM-OLR also revealed a marginally significant
Wording Condition Ļ« Age Group interaction, Wald ā¹2
Ļ 9.24,
df Ļ 4, p Ļ .055. With respect to first prediction, RM-OLRs
performed within each age group revealed that the difference
between the generic and nongeneric conditions was significant for
the 6- to 7-year-olds and the adults (ps Ļ½ .001) and approached
significance for the 4- to 5-year-olds (p Ļ .113; see Figure 1, top).
With respect to the second prediction, RM-OLRs performed
within each age group revealed that the predicted pattern (ge-
neric Ļ āmostā Ļ¾ nongeneric) held up for the 6- to 7-year-olds and
the adults; however, for the 4- to 5-year-olds, the āmostā condition
was not significantly different from either of the other two (ps Ļ¾
.286; see Figure 1, top).
The main RM-OLR identified two other effects. First, the main
effect of property type was significant, Mability Ļ .39 vs.
Mbiological Ļ .48, Wald ā¹2
Ļ 12.65, df Ļ 1, p Ļ½ .001, suggesting
that participants were more likely to essentialize biological fea-
tures than abilities. Second, the main effect of age group was
significant, M4 ā5 Ļ .35 vs. M6 ā7 Ļ .43 vs. Madult Ļ .56, Wald
ā¹2
Ļ 27.53, df Ļ 2, p Ļ½ .001, suggesting that the tendency to
produce essentialized explanations increased with age.
Nonessentialized Explanations
An analogous RM-OLR performed on participantsā nonessen-
tialized explanations provided further evidence for our belief-
proximal account. The crucial main effect of wording condition
was again significant, Wald ā¹2
Ļ 46.71, df Ļ 2, p Ļ½ .001.
Supporting our first prediction, a planned follow-up RM-OLR
revealed that participants in the generic condition produced sig-
nificantly fewer nonessentialized explanations (e.g., she is good at
tooki because she practiced) for the facts in Step 2 than partici-
pants in the nongeneric condition, Mgeneric Ļ .26 vs. Mnongeneric Ļ
.54, p Ļ½ .001.
Supporting our second prediction, planned follow-up RM-OLRs
revealed that the proportion of nonessentialized responses in the
āmostā condition (M Ļ .32) was comparable to that in the generic
condition (M Ļ .26), p Ļ .339, but significantly lower than the
analogous proportion in the nongeneric condition (M Ļ .54), p Ļ½
.001. For example, participants who were told in Step 1 that most
boys or girls are good at a task were relatively reluctant to attribute
the ability of the individual boys or girls in Step 2 to hard work,
which mirrored the behavior of participants in the generic condi-
tion.
The interaction between wording condition and age group in the
main RM-OLR was not significant, Wald ā¹2
Ļ 2.06, df Ļ 4, p Ļ
.725. Despite the nonsignificant interaction, we tested whether the
predicted patterns achieved statistical significance within each age
group. In line with our first prediction, follow-up RM-OLRs
revealed that the differences between the generic and nongeneric
conditions were significant at all age levels (see Figure 1, bottom):
4- to 5-year-olds (p Ļ½ .001), 6- to 7-year-olds (p Ļ½ .001), and
adults (p Ļ .001). In line with our second prediction, follow-up
Figure 1. The mean proportion of essentialized (top) and nonessential-
ized (bottom) explanations, by wording condition and age group. The error
bars represent Ļ® 1 SE.
166 CIMPIAN AND ERICKSON
9. RM-OLRs confirmed that the predicted pattern (generic Ļ āmostā
Ļ½ nongeneric) held up for all three age groups (see Figure 1,
bottom).
Finally, the RM-OLR on all participantsā data revealed significant
main effects of property type, Mability Ļ .49 vs. Mbiological Ļ .29,
Wald ā¹2
Ļ 32.33, df Ļ 1, p Ļ½ .001, and age group, M4ā5 Ļ .38 vs.
M6ā7 Ļ .44 vs. Madult Ļ .34, Wald ā¹2
Ļ 8.98, df Ļ 2, p Ļ .011.
Scope-Restricted Analyses
Although in Step 2 all participants were asked to explain a fact
about a single individual, it is possible that the generic and āmostā
statements in Step 1 may have occasionally caused the participants
in those conditions to think they were asked to explain something
about (most) boys/girls. Such confusions may have introduced
variability in the perceived wording of the questions across the
three conditions, which in turn could have exaggerated the effects
identified above and biased the results in favor of the belief-
proximal account.
To determine how frequent these scope confusions may have
been, we went back and coded which explanations referred to more
than one individual (e.g., āThey do it a lotā [puzzle]; āBecause
girls can get longer hair than boys canā [brain]). The coding
(agreement Ļ 98.5%; ā¬ Ļ .93) revealed that such explanations
occurred on only 13.2% of all trials, suggesting that scope confu-
sions were relatively rare. Moreover, producing a plural explana-
tion in response to a singular question does not necessarily indicate
confusion about the scope of the question. It is perfectly legitimate,
for example, to explain a fact about an individual by appealing to
a feature of the category to which he or she belongs. Consistent
with this idea, plural explanations were in fact more prevalent in
adultsā responses (25%) than they were in childrenās (7%); it
seems implausible to claim that the adults would be more confused
than the children about the scope of our nongeneric questions.
Nevertheless, to test more directly whether these plural expla-
nations had biased our previous conclusions, we also performed
another set of RM-OLR analyses that excluded them; in other
words, we used as our dependent measures the proportions of
nonsyllogistic, nonplural explanations that were essentialized and
nonessentialized. These scope-restricted analyses replicated the
main results described earlier. Most important, the main effect of
wording condition was significant both for the essentialized ex-
planation measure, Mgeneric Ļ .53 vs. Māmostā Ļ .45 vs.
Mnongeneric Ļ .30, Wald ā¹2
Ļ 29.15, df Ļ 2, p Ļ½ .001, and for the
nonessentialized explanation measure, Mgeneric Ļ .27 vs.
Māmostā Ļ .32 vs. Mnongeneric Ļ .56, Wald ā¹2
Ļ 45.42, df Ļ 2, p Ļ½
.001), and follow-up RM-OLRs revealed the patterns of significant
condition differences that would be expected given our two main
predictions.
Conclusion
These results confirmed both predictions of the belief-proximal
account. First, they suggested that the generic beliefs formed on
the basis of exposure to generic statements influence childrenās
causal attributions in subsequent nongeneric contexts that can be
linked with these beliefs. Second, the results suggested that wide-
scope quantified statements can also lead, presumably via the
generic beliefs they license, to essentialized causal attributions in
subsequent nongeneric contexts.
Discussion
Causal relationships are fundamental to human understanding
and human concepts (e.g., Ahn et al., 2000; Gelman, 2003; Mur-
phy & Medin, 1985). By shaping the causal perspective children
adopt with respect to the facts they are learning, the linguistic
distinction between generic and nongeneric statements may have
an important part to play in the development of childrenās lay
theories and of the concepts that are embedded within these
theories. Our goal in this article was to explore the process by
which the generic/nongeneric distinction affects causal attribu-
tions. We proposed that the influence of this linguistic distinction
on childrenās attributions is indirect: The generic or nongeneric
format of the statements that are being explained does not directly
determine (i.e., is not the proximal cause of) childrenās causal
attributions; rather, it is the generic or nongeneric format of the
beliefs relevant to these statements that determines what causal
attributions children will make. To evaluate this proposal, we
formulated and tested two predictions that follow from it and that
distinguish it from the competing language-proximal account, on
which the generic or nongeneric format of the statements that are
being explained is the direct, or proximal, cause of childrenās
attributions.
The First Prediction: Generics Affect Attributions in
Subsequent Nongeneric Contexts
We first predicted that, once formed, a generic belief should be
available in subsequent situations to which it is relevant and should
skew childrenās inferences toward essential causes. In the presence
of a generic belief, an outcome that might typically be construed as
the product of external or mechanistic causes (e.g., she is doing
well because she tried hard; Cimpian & Markman, 2011) is likely
to be understood in terms of the deep traits of the individuals
involved (e.g., she is doing well because, as a girl, she has a special
aptitude for this task).
Our data supported this prediction. Participants who heard ge-
neric statements in Step 1 (e.g., āBoys/Girls are really good at a
game called ātookiāā) were (a) significantly more likely to invoke
deep, essential causes and (b) significantly less likely to invoke
external, mechanistic causes in their explanations for the facts in
Step 2 than were participants who heard nongeneric statements in
Step 1 (e.g., āThereās a boy/girl who is really good at a game called
ātookiāā). These results were strikingly consistent across all the age
groups, with the possible exception of 4- to 5-year-oldsā essential-
ized explanations, where the difference between the generic and
nongeneric conditions only approached significance.
The Second Prediction: Generics Are Not Unique in
Their Effect of Attributions
The second main prediction of our belief-proximal account was
that generic statements should not be unique in their effect on
causal attributions. In particular, we hypothesized that some non-
generic statementsāespecially ones with wide-scope quantifiers
such as āmostā or āallāāmight also lead to inferences about deep,
167GENERICS AND CAUSAL ATTRIBUTIONS
10. essential causes. This prediction was motivated by hints in the
literature that such quantified statements may often give rise to
generic beliefs (Leslie & Gelman, 2011; see also Hollander et al.,
2002; Khemlani et al., 2007). If the information conveyed by these
quantified statements is indeed sufficient to form generic beliefs,
then ultimately these statements should prompt people to essen-
tialize just as generic statements do, since we propose it is the
beliefs (and not the statements themselves) that are the proximal
cause of the attributions.
Our results supported this second prediction as well, in that the
causal attributions of participants exposed to āmostā statements
clustered with those of participants who had heard generic state-
ments in Step 1 and were significantly different from those of
participants who had heard nongeneric statements in Step 1. This
result was consistent across the two measures (essentialized and
nonessentialized explanations) and across the three age groups,
with the same exception of 4- to 5-year-oldsā essentialized expla-
nations (see Figure 1, top). Several ideas about the source of this
developmental difference are discussed in the next section.
We should reiterate that our claim here is not that generic and
āmostā statements have the same meaning. They do not, and there
are many demonstrations that adults draw clear distinctions be-
tween them (e.g., Carlson, 1977; Cimpian, Brandone, & Gelman,
2010; Cimpian, Gelman, & Brandone, 2010). However, the fact
that adults can distinguish between these statement types when
asked to reflect on their meanings does not automatically entail
that the representation of the facts learned from these statements
will be distinct. In fact, the evidence that adults frequently convert
facts learned from āmostā- and āallā-quantified statements to ge-
neric form (Leslie & Gelman, 2011) suggests the opposite may be
true. In sum, our claim is that generic and āmostā statements may
often be equivalent at the level of the beliefs to which they
ultimately give rise. The link from āmostā statements to generic
beliefs might be relatively fragile, however, since such quantified
statements can give rise to generic beliefs only with the aid of
additional inductive inferences or under circumstances that cause
people to fall back on a default kind-based mode of generalization
(Leslie, 2007, 2008).
It is important to note that the subsequent step in this processā
that of generating causal attributions on the basis of the genericity
of the relevant beliefsāmay be similarly variable and sensitive to
contextual factors. Having a generic belief does not automatically
cause an essentialized understanding. Other considerations, such
as the nature of the kinds and properties involved, are also taken
into account in constructing a causal perspective on this fact. For
example, Cimpian and Markman (2011) demonstrated that infer-
ences about the essential nature of the properties introduced via
generic statements were blocked when the categories to which
these statements referred (in this case, boys/girls at a particular
school) were relatively shallow or arbitrary and thus could not
plausibly support attributions to deep, inherent causes.
The Source of the Developmental Differences
Why did 4- to 5-year-oldsā responses deviate from our predic-
tions on the essentialized measure? One interpretation of this result
might be that the belief-proximal mechanism proposed here is not
fully in place in children by the age of 5, such that preschoolersā
causal attributions are less likely to be mediated by their beliefsā
and instead more likely to be driven by the format of the lan-
guageāthan are older childrenās and adultsā attributions. Such a
development might also explain the contrast between this result
and Cimpian and Markmanās (2011) Experiment 1. In that study,
preschoolers were significantly more likely to provide essential-
ized explanations when they were asked about features of kinds
(e.g., why girls are good at gorp) than when they were asked about
features of individuals (e.g., why a particular girl is good at gorp).
That is, when the generic/nongeneric format of the language and of
the beliefs covaried (as in Cimpian & Markmanās study), the
predicted differences in preschoolersā essentialized explanations
were stronger than when the format of the language was held
constant (as in the present study). This is exactly the pattern one
might expect if 4- to 5-year-oldsā causal attributions were less
likely to be mediated by their beliefs than older childrenās attri-
butions.
However, other evidence speaks against this idea. Most impor-
tant, notice the striking similarity between the three age groups
on the nonessentialized explanation measure (see Figure 1, bot-
tom). If the belief-proximal mechanism we hypothesized was still
immature at in children at the age of 5, the differences between
wording conditions should have been equally weak across the
essentialized and nonessentialized measures. In other words, if
young children were less likely to explain the properties of the
individuals in Step 2 (e.g., a girl is good at tooki) in light of the
beliefs induced in Step 1, there is no reason why they should have
produced significantly fewer nonessentialized explanations (e.g.,
she practiced) in the generic and āmostā conditions. Incidentally,
this result also rules out the possibility that 4- to 5-year-olds were
simply unable to comprehend the quantifier āmostā (Papafragou &
Schwarz, 2005ā2006; but see Halberda, Taing, & Lidz, 2008). If
understanding āmostā had been an issue, young childrenās re-
sponses should have looked different from those of older partici-
pants on the nonessentialized measure as well (and they did not).
If there is no developmental change in the process by which
language shapes causal attributions, then what explains 4- to
5-year-oldsā deviation from the predicted pattern on the essential-
ized measure? One clue is provided by the fact that 4- to 5-year-
olds produced overall fewer essentialized explanations than the
older groups (see Figure 1, top). This pattern is consistent with
previous findings that young children are generally less likely than
older children and adults to think of other individualsā behaviors as
stemming from stable traits (Benenson & Dweck, 1986; Lawson &
Kalish, 2006; Rholes & Ruble, 1984; but see Rhodes & Gelman,
2008). Thus, even in the presence of a generic belief, children at
this age may be relatively reluctant to attribute the facts in Step 2
to essential traits of the individuals involved.
Further Questions
Our current proposal raises many additional questions about the
process by which generic and nongeneric statements, and the
beliefs they give rise to, shape causal attributions. For example,
one outstanding question concerns the range of contexts whose
understanding may be influenced by a generic belief. In our study,
the nongeneric fact provided in Step 2 (e.g., āThis boy/girl is also
good at tookiā) was a straightforward example of the generic fact
introduced in Step 1. What about circumstances that are related to
a stored generic belief but that nevertheless differ from it on some
168 CIMPIAN AND ERICKSON
11. important dimension? For example, if children have a generic
belief that boys are good at math, will this belief be accessed when
they see a girl excelling at math? If so, how will it be factored into
their attributions? Will the causal attributions that would likely be
made in the case of individual boys (that there is some talent or gift
underlying their math ability) extend here as well? Or will infer-
ences about underlying talents be restricted to boys and any girlās
ability be explained as the outcome of her sustained efforts? Along
similar lines, we can ask how the attribution process might unfold
when children encounter exceptions to a stored generic belief. For
example, given the generic belief that boys are good at math, how do
they make sense of the performance of individual boys who are not?
Another outstanding question about the process outlined here con-
cerns how long-lasting the effect of a generic statement might be. In
our experiment, participants were asked to generate causal attributions
(Step 2) very soon after exposure to the generic statement (Step 1).
We expect that similar results would hold even if a delay was
introduced between these steps, especially given the evidence that
generic beliefs seem to be encoded robustly in long-term memory
(e.g., Cimpian et al., in press; Gelman & Raman, 2007). However, a
study that actually manipulated this delay would speak more directly
to the persistence of genericsā influence across time.
Finally, in future work, it would be important to adjudicate
between the two hypothesized mechanisms that could lead from
āmostā statements to generic beliefs and thus to essentialized
attributions. Do āmostā-quantified statements give rise to generic
beliefs (a) because they license inductive inferences about the kind
as a whole or (b) because they are more resource-intensive to
process and remember than generic statements, leading people to
distort the quantified meaning of these statements into generic
meaning (Leslie, 2007, 2008; Leslie & Gelman, 2011)? One pos-
sible means of distinguishing between these two mechanisms
would be to reduce the cognitive load involved in processing
āmostā statements (e.g., by providing subjects with extensive
exposure and training before the test).5
If such a reduction was
achievable, the cognitive load account would predict that subjectsā
tendency to default to generic meaning should be diminished,
along with their tendency to essentialize this information. In con-
trast, on the inductive inference account, there is no reason to
expect that this reduction in cognitive load should affect subjectsā
tendency to essentialize the information conveyed by the āmostā
statements; the relevant inductive inferences are equally valid
regardless of how difficult the āmostā statements are to process
and remember.
Conclusion
Generic and nongeneric statements do more than just convey
facts. They also promote particular causal perspectives on these
facts, leading children to understand them either as the result of
deep, inherent, essential factors or as the result of external, mech-
anistic factors, respectively. Our main claim here was that chil-
drenās causal attributions for the facts learned through these state-
ments are determined not by the generic or nongeneric format of
the statements themselves but rather by the generic or nongeneric
format of the beliefs that are relevant to these statements. The two
predictions we derived from this claim were confirmed by our
data. We found, first, that the influence of generic statements is not
limited to just situations in which children are reasoning about
these statements or about the generalizations they express. That is,
generic statementsāor, as we claimed, the generic beliefs they
give rise toāalso shaped childrenās causal attributions in subse-
quent nongeneric contexts that fell under their scope. Thus, the
range of circumstances whose understanding may be influenced by
generic language (via the generic beliefs it induces) is quite ex-
tensive. The second prediction was that wide-scope quantified
statements should also lead to attributions in terms of deep, essen-
tial causes, arguably because these statements often license the
formation of generic beliefs. By demonstrating that generic lan-
guage is not unique in its effects on causal attributions, this result
adds a layer of nuance to our knowledge about the process by
which language shapes understanding. To conclude, this research
illustrates the complex interactions among linguistic input, con-
ceptual knowledge, and causal reasoning, interactions that are at
the core of cognitive development and undoubtedly among its
major driving forces.
5
We thank one of our anonymous reviewers for this suggestion.
References
Abelson, R. P., & Kanouse, D. E. (1966). Subjective acceptance of verbal
generalizations. In S. Feldman (Ed.), Cognitive consistency: Motiva-
tional antecedents and behavioral consequents (pp. 171ā197). New
York, NY: Academic Press.
Ahn, W., Kim, N. S., Lassaline, M. E., & Dennis, M. J. (2000). Causal
status as a determinant of feature centrality. Cognitive Psychology, 41,
361ā416. doi:10.1006/cogp.2000.0741
Ahn, W. K., Kalish, C., Gelman, S. A., Medin, D. L., Luhmann, C., Atran,
S., . . . Shafto, P. (2001). Why essences are essential in the psychology
of concepts. Cognition, 82, 59 ā 69. doi:10.1016/S0010-
0277(01)00145-7
Benenson, J. F., & Dweck, C. S. (1986). The development of trait expla-
nations and self-evaluations in the academic and social domains. Child
Development, 57, 1179ā1187. doi:10.2307/1130441
Bloom, P. (2004). Descartesā baby: How the science of child development
explains what makes us human. New York, NY: Basic.
Carlson, G. N. (1977). Reference to kinds in English. (Unpublished doc-
toral dissertation). University of Massachussetts, Amherst.
Chambers, C. G., Graham, S. A., & Turner, J. N. (2008). When hearsay
trumps evidence: How generic language guides preschoolersā inferences
about unfamiliar things. Language and Cognitive Processes, 23, 749ā
766. doi:10.1080/01690960701786111
Cimpian, A. (2010). The impact of generic language about ability on
childrenās achievement motivation. Developmental Psychology, 46(5),
1333ā1340. doi:10.1037/a0019665
Cimpian, A., Arce, H. C., Markman, E. M., & Dweck, C. S. (2007). Subtle
linguistic cues affect childrenās motivation. Psychological Science, 18,
314ā316. doi:10.1111/j.1467-9280.2007.01896.x
Cimpian, A., Brandone, A. C., & Gelman, S. A. (2010). Generic statements
require little evidence for acceptance but have powerful implications.
Cognitive Science, 34, 1452ā1482. doi:10.1111/j.1551-
6709.2010.01126.x
Cimpian, A., & Cadena, C. (2010). Why are dunkels sticky? Preschoolers
infer functionality and intentional creation for artifact properties learned
from generic language. Cognition, 117, 62ā 68. doi:10.1016/
j.cognition.2010.06.011
Cimpian, A., Gelman, S. A., & Brandone, A. C. (2010). Theory-based
considerations influence the interpretation of generic sentences. Lan-
guage and Cognitive Processes, 25, 261ā276. doi:10.1080/
01690960903025227
169GENERICS AND CAUSAL ATTRIBUTIONS
12. Cimpian, A., & Markman, E. M. (2008). Preschool childrenās use of cues
to generic meaning. Cognition, 107, 19ā53. doi:10.1016/j.cognition
.2007.07.008
Cimpian, A., & Markman, E. M. (2009). Information learned from generic
language becomes central to childrenās biological concepts: Evidence
from their open-ended explanations. Cognition, 113, 14ā25. doi:
10.1016/j.cognition.2009.07.004
Cimpian, A., & Markman, E. M. (2011). The generic/nongeneric distinc-
tion influences how children interpret new information about social
others. Child Development, 82, 471ā 492. doi:10.1111/j.1467-
8624.2010.01525.x
Cimpian, A., Meltzer, T. J., & Markman, E. M. (in press). Preschoolersā
use of morphosyntactic cues to identify generic sentences: Indefinite
singular noun phrases, tense, and aspect. Child Development. doi:
10.1111/j.1467-8624.2011.01615.x
Dweck, C. S. (1999). Self-theories: Their role in motivation, personality,
and development. Philadelphia, PA: Psychology Press.
Dweck, C. S. (2006). Mindset: The new psychology of success. New York,
NY: Random House.
Friedman, A. P. (1972). The headache in history, literature, and legend.
Bulletin of the New York Academy of Medicine, 48, 661ā681.
Galotti, K. M. (1989). Approaches to studying formal and everyday rea-
soning. Psychological Bulletin, 105, 331ā351. doi:10.1037/0033-
2909.105.3.331
Gelman, S. A. (2003). The essential child: Origins of essentialism in
everyday thought. London, United Kingdom: Oxford University Press.
Gelman, S. A. (2009). Learning from others: Childrenās construction of
concepts. Annual Review of Psychology, 60, 115ā140. doi:10.1146/
annurev.psych.59.103006.093659
Gelman, S. A., Chesnick, R. J., & Waxman, S. R. (2005). Motherāchild
conversations about pictures and objects: Referring to categories and
individuals. Child Development, 76, 1129ā1143. doi:10.1111/j.1467-
8624.2005.00876.x-i1
Gelman, S. A., Coley, J. D., Rosengren, K. S., Hartman, E., & Pappas, A.
(1998). Beyond labeling: The role of parental input in the acquisition of
richly structured categories. Monographs of the Society for Research in
Child Development, 63, Serial No. 253.
Gelman, S. A., Goetz, P. J., Sarnecka, B. W., & Flukes, J. (2008). Generic
language in parentchildā conversations. Language Learning and Devel-
opment, 4, 1ā31. doi:10.1080/15475440701542625
Gelman, S. A., & Raman, L. (2003). Preschool children use linguistic form
class and pragmatic cues to interpret generics. Child Development, 74,
308ā325. doi:10.1111/1467-8624.00537
Gelman, S. A., & Raman, L. (2007). This cat has nine lives? Childrenās
memory for genericity in language. Developmental Psychology, 43(5),
1256ā1268. doi:10.1037/0012-1649.43.5.1256
Gelman, S. A., & Tardif, T. (1998). A cross-linguistic comparison of
generic noun phrases in English and Mandarin. Cognition, 66, 215ā248.
doi:10.1016/S0010-0277(98)00021-3
Gelman, S. A., Taylor, M. G., & Nguyen, S. P. (2004). Motherāchild
conversations about gender: Understanding the acquisition of essentialist
beliefs. Monographs of the Society for Research in Child Development,
69, Serial No. 275.
Gelman, S. A., & Wellman, H. M. (1991). Insides and essences: Early
understandings of the nonobvious. Cognition, 38, 213ā244. doi:10.1016/
0010-0277(91)90007-Q
Gopnik, A. (1998). Explanation as orgasm. Minds and Machines, 8, 101ā
118. doi:10.1023/A:1008290415597
Halberda, J., Taing, L., & Lidz, J. (2008). The development of āmostā
comprehension and its potential dependence on counting ability in
preschoolers. Language Learning and Development, 4, 99ā121. doi:
10.1080/15475440801922099
Harris, P. L. (2002). What do children learn from testimony? In P. Car-
ruthers, M. Siegal, & S. Stich (Eds.), The cognitive bases of science (pp.
316ā334). Cambridge, United Kingdom: Cambridge University Press.
doi:10.1017/CBO9780511613517.018
Hollander, M. A., Gelman, S. A., & Star, J. (2002). Childrenās interpreta-
tion of generic noun phrases. Developmental Psychology, 38, 883ā894.
doi:10.1037/0012-1649.38.6.883
Jƶnsson, M. L., & Hampton, J. A. (2006). The inverse conjunction fallacy.
Journal of Memory and Language, 55, 317ā334. doi:10.1016/
j.jml.2006.06.005
Keil, F. C. (1989). Concepts, kinds, and cognitive development. Cam-
bridge, MA: MIT Press.
Kelley, H. H. (1973). The process of causal attribution. American Psychol-
ogist, 28, 107ā128. doi:10.1037/h0034225
Khemlani, S., Leslie, S. J., Glucksberg, S., & Rubio-Fernandez, P. (2007).
Do ducks lay eggs? How people interpret generic assertions. In D. S.
McNamara & J. G. Trafton (Eds.), Proceedings of the 29th Annual
Cognitive Science Society (pp. 395ā400). Austin, TX: Cognitive Science
Society.
Lawson, C. A., & Kalish, C. W. (2006). Inductive inferences across time
and identity: Are category members more alike than single individuals?
Journal of Cognition and Development, 7, 233ā252. doi:10.1207/
s15327647jcd0702_5
Leslie, S. J. (2007). Generics and the structure of the mind. Philosophical
Perspectives, 21, 375ā403. doi:10.1111/j.1520-8583.2007.00138.x
Leslie, S. J. (2008). Generics: Cognition and acquisition. Philosophical
Review, 117, 1ā47. doi:10.1215/00318108-2007-023
Leslie, S. J., & Gelman, S. A. (2011). Quantified statements are recalled as
generics: Evidence from preschool children and adults. Manuscript
submitted for publication.
Lipton, P. (2004). Inference to the best explanation. London, United
Kingdom: Routledge.
Medin, D. L., & Ortony, A. (1989). Psychological essentialism. In S.
Vosniadou, & A. Ortony (Eds.), Similarity and analogical reasoning
(pp. 179ā196). New York, NY: Cambridge University Press. doi:
10.1017/CBO9780511529863.009
Murphy, G. L., & Medin, D. L. (1985). The role of theories in conceptual
coherence. Psychological Review, 92, 289ā316. doi:10.1037/0033-
295X.92.3.289
Papafragou, A., & Schwarz, N. (2005ā2006). āMostā wanted. Language
Acquisition, 13, 207ā251. doi:10.1207/s15327817la1303_3
Rehder, B., & Hastie, R. (2001). Causal knowledge and categories: The
effects of causal beliefs on categorization, induction, and similarity.
Journal of Experimental Psychology: General, 130, 323ā360. doi:
10.1037/0096-3445.130.3.323
Rhodes, M., & Gelman, S. A. (2008). Categories influence predictions
about individual consistency. Child Development, 79, 1270ā1287. doi:
10.1111/j.1467-8624.2008.01188.x
Rholes, W. S., & Ruble, D. N. (1984). Childrenās understanding of dispo-
sitional characteristics of others. Child Development, 55, 550ā560.
doi:10.2307/1129966
Wilson, R. A., & Keil, F. C. (Eds.). (2000). Explanation and cognition.
Cambridge, MA: MIT Press.
Received November 5, 2010
Revision received June 5, 2011
Accepted June 19, 2011 ā ¢
170 CIMPIAN AND ERICKSON